should we use linearised models to calculate fiscal ... · traum and walker (2015). erceg and linde...
TRANSCRIPT
-
EUROPEAN ECONOMY
Economic and Financial Affairs
ISSN 2443-8022 (online)
Jesper Lindé and Mathias Trabandt
DISCUSSION PAPER 064 | JULY 2017
Should We Use Linearised Models to Calculate Fiscal Multipliers?
EUROPEAN ECONOMY
FELLOWSHIP INITIATIVE “Challenges to Integrated Markets”
-
2017 Fellowship Initiative Papers are written by external experts commissioned to write research papers, retaining complete academic independence, contributing to the discussion on economic policy and stimulating debate.
The views expressed in this document are therefore solely those of the author(s) and do not necessarily represent the official views of the European Commission.
Authorised for publication by Mary Veronica Tovšak Pleterski, Director for Investment, Growth and Structural Reforms.
DG ECFIN's Fellowship Initiative 2016-2017 "Challenges to integrated markets" culminates and comes to a successful conclusion with the publication of the fellows' contributed papers in our Discussion paper series. Against the background of increasing strains to economic integration at both the global and the European level, the Initiative has brought together a group of outstanding scholars to re-examine integration challenges at the current juncture and to explore the policy options to address these challenges in a discursive interaction process between the group of fellows and Commission services. The research themes of the fellows have spanned a broad area including topics in the political economy of globalisation and integration, issues of macroeconomic policy making at the zero lower interest rate bound, and market integration challenges not least in view of deepening EMU.
LEGAL NOTICE Neither the European Commission nor any person acting on its behalf may be held responsible for the use which may be made of the information contained in this publication, or for any errors which, despite careful preparation and checking, may appear. This paper exists in English only and can be downloaded from https://ec.europa.eu/info/publications/economic-and-financial-affairs-publications_en.
Europe Direct is a service to help you find answers to your questions about the European Union.
Freephone number (*):
00 800 6 7 8 9 10 11 (*) The information given is free, as are most calls (though some operators, phone boxes or hotels may charge you).
More information on the European Union is available on http://europa.eu.
Luxembourg: Publications Office of the European Union, 2017
KC-BD-17-064-EN-N (online) KC-BD-17-064-EN-C (print) ISBN 978-92-79-64919-6 (online) ISBN 978-92-79-64920-2 (print) doi:10.2765/393098 (online) doi:10.2765/03139 (print)
© European Union, 2017 Reproduction is authorised provided the source is acknowledged. For any use or reproduction of photos or other material that is not under the EU copyright, permission must be sought directly from the copyright holders.
https://ec.europa.eu/info/publications/economic-and-financial-affairs-publications_enhttp://europa.eu.int/citizensrights/signpost/about/index_en.htm#note1#note1http://europa.eu/
-
European Commission Directorate-General for Economic and Financial Affairs
Should We Use Linearised Models to Calculate Fiscal Multipliers? Jesper Lindé and Mathias Trabandt
Abstract We calculate the magnitude of the fiscal spending multiplier in linearised and nonlinear solutions of a New Keynesian model at the zero lower bound. Importantly, the model is amended with real rigidities to simultaneously account for the macroeconomic evidence of a low Phillips curve slope and the microeconomic evidence of frequent price re-optimisation. We show that the nonlinear solution is associated with a much smaller multiplier than the linearised solution in long-lived liquidity traps, and pin down the key features in the model which account for the difference. Our results caution against the common practice of using linearised models to calculate fiscal multipliers in long-lived liquidity traps. JEL Classification: E52, E58. Keywords: Monetary Policy, Fiscal Policy, Liquidity Trap, Zero Lower Bound. Acknowledgements: We are grateful for helpful comments by our discussants Anton Braun, Robert Kollmann, Jinill Kim and Andresa Lagerborg and participants in the Federal Reserve Macro System Committee meeting at the Federal Reserve Bank of Boston in November 2013, the ECB/EACBN/Atlanta Fed conference "Nonlinearities in macroeconomics and finance in the light of crises" hosted by the European Central Bank in December 2014, the Korean Economic Association and Social Science conference "Recent Issues in Monetary Economics" in Seoul October 2016, and the European University Institute workshop "Economic Policy Challenges" in Florence November 2016, respectively. In addition, we have benefitted from comments from participants in seminars at Norges Bank, Oslo University, Sveriges Riksbank, University of Vienna, North Carolina State University, Central Bank of Finland, University of Hamburg, Goethe University Frankfurt am Main, European Commission, Halle University, Central Bank of Korea and Humboldt Universität Berlin, Bank of England, Queen Mary University and Atlanta Fed. Special thanks to Svetlana Chekmasova, Sher Singh and Mazi Kazemi for outstanding research assistance. Part of this paper was written when the authors were employees in the International Finance Division at the Federal Reserve Board (Lindé and Trabandt), International Monetary Fund (Lindé) and the DG-ECFIN 2016/17 Fellowship Initiative (Trabandt). The authors are grateful to these institutions for the support of this project. The views expressed in this paper are solely the responsibility of the authors and should not be interpreted as reflecting the views of Sveriges Riksbank. First version: October 25, 2013 This version: April 30, 2017 Contacts: Jesper Lindé, Research Division, Sveriges Riksbank, SE-103 37 Stockholm, Sweden, e-mail: [email protected]. Mathias Trabandt, Freie Universität Berlin, School of Business and Economics, Chair of Macroeconomics, Boltzmannstrasse 20, 14195 Berlin, Germany, e-mail: [email protected].
EUROPEAN ECONOMY Discussion Paper 064
mailto:[email protected]
-
CONTENTS
1. Introduction ...................................................................................................................................................... 2
2. A Stylised New Keynesian Model ............................................................................................................. 5
2.1. Model ....................................................................................................................................................... 5
2.1.1. Households ................................................................................................................................................... 5
2.1.2. Firms and Price Setting ............................................................................................................................... 6
2.1.3. Monetary and Fiscal Policies ..................................................................................................................... 9
2.1.4. Aggregate Resource Constraint .............................................................................................................. 9
2.2. Parametrisation .................................................................................................................................... 10
2.3. Solving the Model ................................................................................................................................ 12
3. Results .................................................................................................................................... 13
3.1. Construction of Baseline Scenarios .................................................................................................. 14
3.2. Marginal Fiscal Multipliers ................................................................................................................... 15
4. Robustness analysis ..................................................................................................................... 19
4.1. Dixit-Stiglitz vs. Kimball ......................................................................................................................... 19
4.2. Indexation of non-optimising firms .................................................................................................... 20
5. Conclusions .................................................................................................................................................... 21
REFERENCES …………………………………………………………….……………………..…..……… 22
TABLES AND FIGURES .……………………………………………….…………………..……..……… 26
APPENDIX A - ADDITIONAL RESULTS FOR THE STYLISED MODEL A.1. The Log-linearised Stylised Model ................................................................................................................. 32
A.2. Sensitivity Analysis in the Simple Model ....................................................................................................... 33
-
1. Introduction
The magnitude of the fiscal spending multiplier is a classic subject in macroeconomics. To calculate
the magnitude of the multiplier, economists typically employ a linearized version of their actual
nonlinear model. Does linearizing the nonlinear model matter for the conclusions about the mul-
tiplier? We document this may be the case, especially in long-lived liquidity traps. When interest
rates are expected to be constrained by the zero (or effective) lower bound for a protracted time
period, the nonlinear solution suggests a much smaller multiplier than the linearized solution of the
same model.
The financial crisis and “Great Recession”have revived interest in the magnitude of the fiscal
spending multiplier. A quickly growing literature suggests that the fiscal spending multiplier can
be very large when nominal interest rates are expected to be constrained by the zero (or effective)
lower bound (ZLB) for a prolonged period, see e.g. Eggertsson (2010), Davig and Leeper (2011),
Christiano, Eichenbaum and Rebelo (2011), Woodford (2011), Coenen et al. (2012) and Leeper,
Traum and Walker (2015). Erceg and Linde (2014) show that in a long-lived liquidity trap fiscal
stimulus can come at low cost to the treasury and even be a “fiscal free lunch”. Conversely, the
results of the above literature suggest that it is diffi cult to reduce government debt in the short-run
through aggressive government spending cuts in long-lived liquidity traps: fiscal consolidation can
in fact be self-defeating in such a situation.
Importantly, the bulk of the existing literature analyzes fiscal multipliers in models where all
equilibrium equation have been linearized around the steady state, except for the ZLB constraint
on the monetary policy rule. Implicit in the linearization procedure is the assumption that the
linearized solution is accurate even far away from the steady state. However, recent work by
Boneva, Braun, and Waki (2016) suggests that linearization produces severely misleading results
at the zero lower bound. Essentially, Boneva et al. argue that extrapolating decision rules far away
from the steady state is invalid.
Our paper provides a positive analysis of the effect of spending-based fiscal stimulus on output
and government debt using a fully nonlinear model. We compare the fiscal spending multipliers for
output and government debt of the nonlinear and linearized solution as function of the liquidity
trap duration. Moreover, our framework allows us to pin down the key features which account
for the difference between the multiplier schedule for the nonlinear and linearized solutions of the
model.
2
-
The model environment is a variant of the canonical New Keynesian DSGE model of Woodford
(2003). This model features monopolistic competition and Calvo sticky prices and the central bank
follows a Taylor rule subject to the ZLB constraint on nominal rates.1 A distinct difference to the
existing literature which uses variations of the standard New Keynesian model —for instance the
recent work by Boneva, Braun and Waki (2016), Christiano, Eichenbaum and Johannsen (2016),
Eggertsson and Singh (2016), Fernandez-Villaverde et al. (2015) and Nakata (2015) — is that
we consider a framework with real rigidities. Specifically, we introduce real rigidities through
the Kimball (1995) state-dependent demand elasticity which allows our model to simultaneously
account for the macroeconomic evidence of a low linearized Phillips curve slope (0.01) and the
microeconomic evidence of frequent price re-optimization (3-4 quarters).2 Due to its ability to the
ease the tensions between the macro- and microeconomic evidence on price setting, the Kimball
aggregator has gained traction in New Keynesian models and it is for example used in the workhorse
Smets and Wouters (2007) model.
Our main results are as follows. First, we show that the nonlinear solution is associated with
a much smaller fiscal spending multiplier than the linearized solution in long-lived liquidity traps.
More precisely, when the ZLB is expected to bind for 3 years, the nonlinear solution implies a
multiplier of about 2/3 while the linearized solution of the same model implies a multiplier slightly
above 2. As the multiplier in our benchmark model with real rigidities equals 1/3 in normal times
when the ZLB does not bind, the nonlinear solution implies roughly a doubling of the multiplier
in a long-lived trap, whereas the linearized solution implies that the multiplier is elevated seven
times.
What accounts for the large difference between the nonlinear and linearized solutions in a
prolonged liquidity trap? We document that the difference can almost entirely be accounted for
by the nonlinearities in the price setting block of the model —the Phillips curve. Key here is the
nonlinearity implied by the Kimball aggregator. The Kimball aggregator implies that the demand
elasticity for intermediate goods is state-dependent, i.e. the firms’demand elasticity is an increasing
function of its relative price. In short, the demand curve is quasi-kinked. While the fully nonlinear
model takes this state-dependency explicitly into account, a linear approximation replaces that
nonlinearity by a linear function. Put differently, linearization replaces the quasi-kinked demand
1 We rule out the well-known problems associated with steady state multiplicity emphasized by Benhabib, Schmitt-Grohe and Uribe (2001) by restricting our attention to the steady state with a positive inflation rate.
2 There is also a recent literature which studies models where the effects of government spending is state dependent(i.e. differs in booms and recessions even absent zero lower bound considerations due to labor market frictions), seee.g. Michaillat (2014), Rendahl (2016) and Roulleau-Pasdeloup (2016). This paper does not address this literature.
3
-
curve with a linear function.3 Intuitively, in a deep recession that triggers the ZLB to bind for a
long time, the Kimball aggregator carries the implication that firms do not find it attractive to cut
their prices much since that reduces the demand elasticity and thereby does not crowd in more
demand. With more fiscal spending in such a situation, firms also find it less attractive to increase
their prices. Thus —with policy rates stuck at zero —aggregate inflation increases only little and
therefore the real interest rate falls by little: the multiplier does not increase to the same extent
with the duration of the ZLB. When the model is linearized, the response of aggregate inflation is
notably stronger due to the nature of a linear approximation of a quasi-kinked demand curve at
the steady state with no dispersion. Hence, the drop in the real interest rate is elevated following
a spending hike and the multiplier is magnified. The bottom line: the linearized version of the
model exaggerates the rise in expected and actual inflation due to a sizable approximation error
and thereby elevates the magnitude of the fiscal multiplier in long-lived liquidity traps.4
Our results have potentially important implications for the scope of fiscal stimulus to be self-
financing, and the extent to which fiscal consolidations can be self-defeating. In the nonlinear
model, fiscal stimulus is never a “free lunch”as in Erceg and Lindé (2014); and conversely, fiscal
consolidations are never self-defeating. The linearized solution arrives at the opposite conclusions:
fiscal stimulus can be self-financing in a suffi ciently long-lived liquidity trap and fiscal consolidations
can be self-defeating. However, although these findings cast doubt on the findings of the existing
literature on the fiscal implications of stimulus, it should be recalled that we are considering a model
environment where the fiscal output multiplier is small in normal times (1/3 as mentioned earlier).
Had we instead considered a model in which the multiplier were in the mid-range of the empirical
evidence (i.e. a multiplier slightly less than unity) when monetary policy is unconstrained, it is
possible that the multiplier could be magnified suffi ciently in a long-lived liquidity trap to obtain
a “fiscal free lunch” for a transient hike in spending.5 We elaborate further on this issue in the
conclusions.
Our paper is related to Boneva, Braun andWaki (2016), Christiano, Eichenbaum and Johannsen
(2016), Fernandez-Villaverde et al. (2015), Nakata (2015) and Eggertsson and Singh (2016). Im-
portantly, none of the above papers considers the case of a Kimball (1995) aggregator. Boneva,
3 It is well known that in a linearized model, the Kimball (1995) and Dixit-Stiglitz (1977) aggregator —the latterfeaturing a constant demand elasticity —are observationally equivalent up to a factor of proportionality.
4 The small rise in inflation expectations is consistent with Dupor and Li (2015), who argue that expected inflationreacted little to spending shocks in the United States during the Great Recession.
5 A large empirical literature has examined the effects of government spending shocks, mainly focusing on thepost-WWII pre-financial crisis period when monetary policy had latitude to adjust interest rates. The bulk of thisresearch suggests a government spending multiplier in the range of 0.5 to somewhat above unity (1.5). See e.g. Hall(2009), Ramey (2011), Blanchard, Erceg and Lindé (2016) and the references therein.
4
-
Braun and Waki (2016) report that the multiplier is smaller in a fully nonlinear model. Their model
features a Dixit-Stiglitz aggregator. Eggertsson and Singh (2016) report that the multipliers of the
nonlinear and linearized model differ only very little. Their model features a Dixit-Stiglitz aggre-
gator and assumes firms-specific labor markets, implying that price dispersion is irrelevant for the
nonlinear model dynamics. By contrast, our analysis shows how important these assumptions are:
moving to the frequently used Kimball aggregator and allowing for price dispersion alters the con-
clusions about the multiplier substantially. Nakata (2015) and Fernández-Villaverde et al. (2015)
show that shock uncertainty may have potentially important implications for equilibrium dynamics.
Even so, robustness analysis shows that allowing for shock uncertainty have very limited impact on
our results when the degree of price adjustment (and thus the extent to which how quickly expected
inflation adjusts) is calibrated to fit the macroevidence on the slope of Phillips curve. Christiano,
Eichenbaum and Johannsen (2016) analyze multiplicity of equilibria in a nonlinear New Keynesian
model. They document that there is a unique stable-under-learning rational expectations equilib-
rium in their model and that all other equilibriums are not stable under learning.
The remaining of the paper is organized as follows. Section 2 presents the small scale New
Keynesian model and Section 3 the results. Section 4 examines the robustness of the results to
various perturbations of the model. Section 5 concludes.
2. A Stylized New Keynesian Model
The simple model we study is very similar to the one developed Erceg and Linde (2014), which in
turn builds on the baseline Eggertsson and Woodford (2003) model with the exception that it allows
for real effects of price distortions by dropping the assumption that labor cannot be reallocated
between different firms (or industries). We deviate from Erceg and Linde (2014) in two ways; first by
allowing for a Kimball (1995) aggregator (with the standard Dixit and Stiglitz (1977) specification
as a special case), and second, by including a discount factor (or savings) shock. Below, we outline
the model and its key nonlinear equations. In the Appendix A, we describe the linearized version.
2.1. Model
2.1.1. Households
The utility functional for the representative household is
5
-
max{Ct,Nt,Bt}∞t=0
E0
∞∑t=0
βtςt
{log (Ct − Cνt)−
N1+χt1 + χ
}(1)
where the discount factor β satisfies 0 < β < 1 and is subject to an exogenous shock ςt. As in
Erceg and Lindé (2014), the utility function depends on the household’s current consumption Ct
as deviation from a “reference level”Cνt+j . The exogenous consumption taste shock νt raises the
reference level and marginal utility of consumption. The utility function also depends negatively
on hours worked Nt.
The household’s budget constraint in period t states that its expenditure on goods and net
purchases of (zero-coupon) government bonds BG,t must equal its disposable income:
PtCt +BG,t = (1− τN )WtNt + (1 + it−1)BG,t−1 − Tt + Γt (2)
Thus, the household purchases the final consumption good at price Pt. The household is subject
to a constant distortionary labor income tax τN and earns after-tax labor income (1− τN )WtNt.
The household pays lump-sum taxes net of transfers Tt and receives a proportional share of the
profits Γt of all intermediate firms.
Utility maximization yields the standard consumption Euler equation
1 = βδtEt
{(1 + it)
1 + πt+1
Ct − CνtCt+1 − Cνt+1
}, (3)
where we have defined
δt ≡Etςt+1ςt
, (4)
and introduced the notation 1 +πt+1 = Pt+1/Pt. We also have the following labor supply schedule:
Nχt =1− τNCt − Cνt
WtPt. (5)
Equations (3) and (5) are the key equations for the household side of the model.
2.1.2. Firms and Price Setting
Final Goods Production The single final output good Yt is produced using a continuum of differen-
tiated intermediate goods Yt(f). Following Kimball (1995), the technology for transforming these
intermediate goods into the final output good is∫ 10G
(Yt (f)
Yt
)df = 1. (6)
6
-
As in Dotsey and King (2005) and Levin, Lopez-Salido and Yun (2007), we assume that G (·) is
given by the following strictly concave and increasing function:
G
(Yt (f)
Yt
)=
ω
1 + ψ
[(1 + ψ) Yt(f)Yt − ψ
] 1ω −
[ω
1 + ψ− 1], (7)
where ψ = (1−φp)�pφp
and ω =φp−(φp−1)�p1−(φp−1)�p
. Here φp ≥ 1 denotes the gross markup of the intermediate
firms. The parameter �p governs the degree of curvature of the intermediate firm’s demand curve,
and in Figure 1 we show how relative demand is affected by the relative price under alternative
assumptions about �p (and thus ψ) for given φp.6 When �p = 0, the demand curve exhibits constant
elasticity as under the standard Dixit-Stiglitz aggregator, implying a linear relationship between
relative demand and relative prices (ψ = 0 in Figure 1). When �p is positive —as in Smets and
Wouters (2007) — the firm’s instead face a quasi-kinked demand curve, implying that a drop in
its relative price only stimulates a small increase in demand. On the other hand, a rise in its
relative price generates a large fall in demand. Relative to the standard Dixit-Stiglitz aggregator,
this introduces more strategic complementarity in price setting which causes intermediate firms to
adjust prices less to a given change in marginal cost, especially when �p is high (ψ = −8 in Figure
1). Finally, we notice that G(1) = 1, implying constant returns to scale when all intermediate firms
produce the same amount.
Firms that produce the final output good are perfectly competitive in both product and factor
markets. Thus, final goods producers minimize the cost of producing a given quantity of the output
index Yt, taking as given the price Pt (f) of each intermediate good Yt(f). Moreover, final goods
producers sell units of the final output good at a price Pt, and hence solve the following problem:
max{Yt,Yt(f)}
PtYt −∫ 1
0Pt (f)Yt (f) df (8)
subject to the constraint (6). The first order conditions can be written as
Yt(f)Yt
= 11+ψ
([Pt(f)Pt
1Λpt
]−φp−(φp−1)�pφp−1 + ψ
), (9)
PtΛpt =
[∫Pt (f)
− 1−(φp−1)�pφp−1 df
]− φp−11−(φp−1)�p
,
Λpt = 1 + ψ − ψ∫
Pt(f)Pt
df,
6 The figure is taken from Levin, Lopez-Salido and Yun (2007), and the mapping between ψ and �p is given by�p =
−φpψ(φp−1)
. A value of ψ = −8 thus implies that �p equals 88 when the gross markup φp equals 1.1.
7
-
where Λpt denotes the Lagrange multiplier on the aggregator constraint (7). Note that for �p = 0
(and thus ψ = 0), Λpt = 1 ∀t and the first-order conditions in (9) simplifies to the usual Dixit and
Stiglitz (1977) expressions
Yt (f)
Yt=
[Pt (f)
Pt
]− φpφp−1
, Pt =
[∫Pt (f)
11−φp df
]1−φpIntermediate Goods Production A continuum of intermediate goods Yt(f) for f ∈ [0, 1] is produced
by monopolistically competitive firms, each of which produces a single differentiated good. Each
intermediate goods producer faces a demand schedule from the final goods firms through the solu-
tion to the problem in eq. (8) that varies inversely with its output price Pt (f) and directly with
aggregate demand Yt.
Aggregate capital (K) is assumed to be fixed, so that aggregate production of the intermediate
good firm is given by
Yt (f) = K (f)αNt (f)
1−α . (10)
Despite the fixed aggregate stock K ≡∫K (f) df , shares of it can be freely allocated across the f
firms, implying that real marginal cost, MCt(f)/Pt is identical across firms and equal to
MCtPt≡ Wt/PtMPLt
=Wt/Pt
(1− α)KαNt−α, (11)
where Nt =∫Nt (f) df .
The prices of the intermediate goods are determined by Calvo-Yun (1996) style staggered nom-
inal contracts. In each period, each firm f faces a constant probability, 1 − ξp, of being able to
re-optimize its price Pt(f). The probability that any firm receives a signal to reset its price is as-
sumed to be independent of the time that it last reset its price. If a firm is not allowed to optimize
its price in a given period, it adjusts its price according to the following formula
P̃t = (1 + π)Pt−1, (12)
where π is the steady-state (net) inflation rate and P̃t is the updated price.
Given Calvo-style pricing frictions, firm f that is allowed to re-optimize its price, P optt (f), solves
the following problem
maxP optt (f)
Et∞∑j=0
(βξp)jςt+jΛt,t+j
[(1 + π)j P optt (f)−MCt+j
]Yt+j (f)
where Λt,t+j is the stochastic discount factor (the conditional value of future profits in utility units,
recalling that the household is the owner of the firms), and demand Yt+j (f) from the final goods
firms is given by the equations in (9).
8
-
2.1.3. Monetary and Fiscal Policies
The evolution of nominal government debt is determined by the following equation
BG,t = (1 + it−1)BG,t−1 + PtGt − τNWtNt − Tt (13)
where Gt denotes real government expenditures on the final good Yt. Following the convention in
the literature on fiscal multipliers, we assuming that lump-sum taxes stabilize government debt
as share of nominal trend GDP, bG,t ≡ BG,tP̄tY . Specifically, we follow Erceg and Linde (2014) and
assume that net lump-sum taxes as share of nominal trend GDP, τ t ≡ TtP̄tY , follow the simple rule:
τ t − τ = ϕb (bG,t−1 − bG) , (14)
where variables without time subscript denote steady state. Finally, government spending, gy,t ≡ GtYis exogenous.
Turning to the central bank, it is assumed to adhere to a Taylor-type policy rule that is subject
to the zero lower bound:
1 + it = max
(1, (1 + i)
[1 + πt1 + π
]γπ [ YtY pott
]γx)(15)
where Y pott denotes the level of output that would prevail if prices were flexible, and i the steady-
state (net) nominal interest rate, which is given by r + π where r ≡ 1/β − 1. In the linearized
model, (15) is written
it = max (0, i+ γπ (πt − π) + γxxt) (16)
where xt ≡ ln(Yt/Y
pott
)is the model-consistent output gap.
2.1.4. Aggregate Resource Constraint
We now turn to discuss the derivation of the aggregate resource constraint. Let Y sumt denote the
unweighted average (sum) of output for each firm f , i.e.
Y sumt =
∫ 10Yt(f)df.
which from (10) and the observation that all firms have the same capital-labor ratio can be rewritten
as
Y sumt =
∫ (K (f)
Nt(f)
)αNt(f)df
=
(K
Nt
)α ∫Nt(f)df (17)
= KαN1−αt
9
-
Recalling that Yt+j (f) is given from (9), it follows that
Y sumt = Yt
∫ 10
11+ψ
([Pt(f)Pt
1Λpt
]−φp−(φp−1)�pφp−1 + ψ
)df,
or equivalently, using (17):
Yt = (p∗t )−1KαN1−αt , (18)
where
p∗t ≡∫ 1
0
φpφp−(φp−1)�p
([Pt(f)Pt
1Λpt
]−φp−(φp−1)�pφp−1 + ψ
)df.
In a technical appendix, available upon request, we show how to develop a recursive formulation of
the sticky price distortion term p∗t .
Now, because actual output Yt is what is available for private consumption and government
spending purposes, it follows that:
Ct +Gt︸ ︷︷ ︸≡Yt
≤ (p∗t )−1KαNt
1−α︸ ︷︷ ︸≡Y sumt
. (19)
The sticky price distortion introduces a wedge between input use and the output available for
private and government consumption.7 Even so, this term vanishes in the log-linearized version of
the model.
2.2. Parameterization
Our benchmark calibration − essentially adopted from Erceg and Linde (2014) − is fairly standard
at a quarterly frequency. We set the discount factor β = 0.995, and the steady state net inflation
rate π = .005; this implies a steady state interest rate of i = .01 (i.e., four percent at an
annualized rate). We set the intertemporal substitution elasticity σ = 1 (log utility), the capital
share parameter α = 0.3, the Frisch elasticity of labor supply 1χ = 0.4, and the steady state value for
the consumption taste shock ν = 0.01.8 Three parameters determine the direct sensitivity of prices
to marginal costs: the gross markup φp, the stickiness parameter ξp and the Kimball parameter
�p. We have direct evidence on two of these —φp and ξp. When it comes the latter, a large body
of microeconomic evidence, see e.g. Klenow and Malin (2010) and Nakamura and Steinsson (2012)
and the references therein, suggest that firms change their prices rather frequently, on average
7 As the economy is assumed to be endowned with a fixed aggregate capital stock K which does not depreciate, noresources is devoted to investment. An alternative formulation would have embodied a constant capital depreciationrate in which case output would have been used for Ct, I and Gt.
8 By setting the steady value of the consumption taste shock to a small value, we ensure that the dynamics forthe other shocks are roughly invariant to the presence of −Cνt in the period consumption utility function.
10
-
somewhat more often than once a year. Based on this micro evidence, we set ξp = 0.667, implying
an average price contract duration of 3 quarters ( 11−0.667). For the gross markup, we set φp = 1.2 as
a compromise between the low estimate of φp in Altig et al. (2011) and the higher estimated value
by Smets and Wouters (2007). In addition, this is the estimated value by Christiano, Eichenbaum
and Evans (2005) and Christiano, Trabandt and Walentin (2010), and used by e.g. Levin, Lopez-
Salido and Yun (2007) and Eggertsson and Singh (2016). To pin down �p, our starting point is the
loglinearized New Keynesian Phillips Curve
π̂t = βEtπ̂t+1 + κmcm̂ct, (20)
which obtains in our model where m̂ct denotes marginal cost as log-deviation from its steady state
value. The parameter κmc, i.e. the slope of the Phillips curve, is given by
κmc ≡(1−ξp)(1−βξp)
ξp
11+(φp−1)�p
. (21)
The macroeconomic evidence suggest that the sensitivity of aggregate inflation to variations in
marginal cost is very low, see e.g. Altig et al. (2011). To capture this, we adopt a value for
�p so that the slope of the Phillips curve (κmc) − given our adopted values for β, ξp and φp−
equals 0.012.9 This calibration allows us to match both the micro- and macroevidence on price
setting behavior and is aimed at capturing the resilience of core inflation, and measures of expected
inflation, during the recent global recession.
We assume a government debt to annualized output ratio of 0.6 (consistent with U.S. pre-crisis
federal debt level). We set government consumption as a share of output gy = 0.2. Further, we
set net lump-sum taxes τ = 0 in steady state. The above assumptions imply a steady state labor
income tax τN = 0.33. The parameter ϕ in the tax rule (14) is set equal to 0.01, which implies
that the contribution of lump-sum taxes to the response of government debt is negligible in the first
couple of years following a shock (so that almost all variation in tax revenues reflect fluctuations
in labor tax revenues). For monetary policy, we use the standard Taylor (1993) rule parameters
γπ = 1.5 and γx = .125.
In order to facilitate comparison between the nonlinear and linear model, we specify processes
for the exogenous shocks such that there is no loss in precision due to an approximation. In
particular, the preference, discount and government spending shocks are assumed to follow AR(1)
9 The median estimates of the Phillips Curve slope in recent empirical studies by e.g. Adolfson et al (2005), Altiget al. (2011), Galí and Gertler (1999), Galí, Gertler and López-Salido (2001), Lindé (2005), and Smets and Wouters(2003, 2007) are in the range of 0.009− .014.
11
-
processes:
gy,t − gy = ρg (gy,t−1 − gy) + εg,t,
νt − ν = ρν (νt−1 − ν) + σν,t, (22)
δt − δ = ρδ (δt−1 − δ) + σδεδ,t,
where δ = 1. Our baseline parameterization of these processes adopts a persistence coeffi cient of
0.95, so that ρν = ρg = ρδ = 0.95 in (22). But following some prominent papers in the literature on
fiscal multipliers, we also investigate the sensitivity of our results when the processes are assumed
to be moving average (MA) processes. Those results are reported in Appendix A.
2.3. Solving the Model
We compute the linearized and nonlinear solutions using the Fair and Taylor (1983) method. This
method imposes certainty equivalence on the nonlinear model, just as the linearized solution does
by definition. In other words, the Fair and Taylor solution algorithm traces out the implications of
not linearizing the equilibrium equations for the resulting multiplier without shock uncertainty. An
alternative approach would have been to compute solutions where uncertainty about future shock
realizations matters for the dynamics of the economy following for instance Adam and Billi (2006,
2007) within a linearized framework and Fernández-Villaverde et al. (2015) and Gust, Herbst,
López-Salido and Smith (2016) within a nonlinear framework. These authors have shown that
allowing for future shock uncertainty can potentially have important implications for equilibrium
dynamics, especially when inflation expectations are less well anchored because the conduct of
monetary policy is far from non-optimal and prices are quick to adjust. We nevertheless confine
ourself to study perfect foresight simulations for the following three reasons. First, because much
of the existing literature have in fact used a perfect foresight approach, retaining this approach
allows us to parse out the effects of going from a linearized to a nonlinear framework. Second, the
high degree of real rigidities we introduce in order to fit the micro- and macroeconomic evidence
implies that expected inflation adjusts slowly, which in turn means that the impact of future shock
uncertainty is modest. Even so, allowing for shock uncertainty only strengthens our finding of a
pronounced difference between the linearized and nonlinear solutions in a long-lived trap as the
responses in the linearized equilibrium is amplified relatively more than the nonlinear solution by
future shock uncertainty.10 Third, the perfect foresight assumption allows us to readily study the10 We have verified this by comparing the decision rules under perfect foresight with the decision rules obtained
when allowing for shock uncertainty (calibrating the variance of the shocks in the model so that the probability is
12
-
robustness in a larger scale model with many state variables. So far, the solution algorithms used
to solve models with shock uncertainty have typically not been applied to models with more than
4-5 state variables.11
To solve the model, we feed the relevant equations in the nonlinear and log-linearized versions
of the model to Dynare. Dynare is a pre-processor and a collection of MATLAB routines which can
solve nonlinear models with forward looking variables, and the details about the implementation
of the algorithm used can be found in Juillard (1996). We use the perfect foresight simulation
algorithm implemented in Dynare using the ‘simul’command.12 The algorithm can easily handle
the ZLB constraint: one just writes the Taylor rule including the max operator in the model
equations, and the solution algorithm reliably calculates the model solution is fractions of a second.
Thus, apart from gaining intuition about the mechanisms embedded into the models, there is no
need anymore to linearize models in order to solve and simulate them.
For the linearized model, we used the algorithm outlined in Hebden, Linde and Svensson (2012)
to check for uniqueness of the local equilibrium associated with a positive steady state inflation
rate. As noted earlier, we rule the well-known problems associated with steady state multiplicity
emphasized by Benhabib, Schmitt-Grohe and Uribe (2001) by restricting our attention to the steady
state with a positive inflation rate. However, for the nonlinear version of the model we cannot rule
out the possibility that there exists other solutions in addition to the one found by Dynare, but note
that the work by Christiano, Eichenbaum and Johannsen (2016) suggest that alternative solutions
may not be relevant (i.e. not stable under learning).
3. Results
In this section, we report the benchmark results. As mentioned earlier, our aim is to compare
spending multipliers in linearized and nonlinear versions of the model economy. Specifically, we
seek to characterize how the difference between the multiplier in the linear and nonlinear frameworks
varies with the expected duration of the liquidity trap. We start out by reporting how we construct
the baseline scenarios and then report the marginal fiscal multipliers.
hitting the ZLB is 0.2 percent each quarter of the stochastic simulations with the model, in line with Post WWIIU.S. evidence).11 A recent paper by Judd, Maliar and Maliar (2011) provides a promising avenue to compute the stochastic
solution of larger scale models effi ciently.12 The solution algorithm implemented in Dynare’s simul command is the method developed in Fair and Taylor
(1983).
13
-
3.1. Construction of Baseline Scenarios
To construct a baseline where the interest rate is bounded at zero for ZLBDUR = 1, 2, 3, ..., T
periods, we follow the previous fiscal multiplier literature (e.g. Christiano, Eichenbaum and Rebelo,
2011) and assume that the economy is hit by a large adverse shock that triggers a deep recession
and drives interest rates to zero. The longer the expected liquidity trap duration (i.e. the larger
value of ZLBDUR) we want to consider, the larger the adverse shock has to be. The particular
shock we consider is a negative consumption taste shock νt (see the equations 1 and 22) following
Erceg and Linde (2014).13
To provide clarity on how we pick the shock sizes, Figure 2 reports the linear and nonlinear
solutions for the same negative taste shock (depicted in the bottom right panel). The economy
is in the deterministic steady state in period 0, and then the shock hits the economy in period 1.
As is evident from Figure 2, the same-sized shock has a rather different impact on the economy
depending on whether the model is linearized or solved in its original nonlinear form. For instance,
we see from panel 3 that while the nominal interest rate is bounded by zero from periods 1 to 8
in the linearized model, the same-sized consumption demand shock (panel 9) only generates a two
quarter trap in the nonlinear model. Hence, we need to subject the nonlinear model to a more
negative consumption demand shock − as shown by the red dotted line in panel 9 in Figure 3 −
to generate ZLBDUR = 8 for the interest rate (panel 3).14
Important insights about the differences between the linearized and nonlinear solutions can be
gained from Figures 2 and 3. Starting with Figure 2, we see from the fifth panel that the drop
in the potential real rate is about the same in both models. Still, the linearized model generates
a much longer liquidity trap because inflation and expected inflation falls much more (panel 2),
which in turn causes the actual real interest rate (panel 4) to rise much more initially. The larger
initial rise in the actual real interest rate, and thus the gap between the actual and potential real
rates, triggers a larger fall in the output gap (panel 1) and consequently real GDP falls more in the
linearized model as well (because the impact on potential GDP is about the same, as implied by
the similarity of the potential real interest rate response).
Turning to Figure 3, we first note from the third panel that the paths for the policy rate are
13 In Appendix A, we present results when the recession is instead assumed to be triggered by the discount factorshock δt that was used in the seminal papers by Eggertsson and Woodford (2003), and Christiano, Eichenbaum andRebelo (2011). For the linearized solution, the results are invariant w.r.t. the choice of the baseline shock (see Ercegand Lindé, 2014, for proof). And for the nonlinear solution, Figure A.1 shows that the multiplier schedules are nearlyidentical.14 Figure 3 also depicts a third line (“Nonlinear model with linear NKPC and Res. Con.”), which we will discuss
further in Section 3.2.
14
-
bounded at zero for 8 quarters and display a very similar path upon exit from the liquidity trap.
Moreover, panel 9 shows that it takes a much larger adverse consumption demand shock in the
nonlinear model to trigger a liquidity trap of the same expected duration as in the linearized model.
This implies that the drop in the potential real rate and real GDP (panels 5 and 7) is much more
severe in the nonlinear model. Even so, and perhaps most important, we see that inflation − panel 2 − falls substantially less in the nonlinear model. This suggests that the difference between the
linearized and nonlinear solutions to a large extent is driven by the pricing block of the model.
3.2. Marginal Fiscal Multipliers
As previously noted, we are seeking to compare fiscal multipliers in liquidity traps of same expected
duration in the linearized and nonlinear frameworks. Accordingly, we allow for differently sized
shocks in the linearized and nonlinear models so that each model variant generates a liquidity
trap with the same expected duration ZLBDUR = 1, 2, 3, ..., T . Let{Blineart
(σlinearν,i
)}Tt=1
and{Bnonlint
(σnonlinν,i
)}Tt=1
denote vectors with simulated variables in the linear and nonlinear models
in periods t = 1, 2, ..., T , respectively. This notation reflects that the innovations, σν , to the
consumption demand shock νt, in eq. (22) are set so that
σlinearν,i ⇒ ZLBDUR = i,
and
σnonlinν,i ⇒ ZLBDUR = i,
where we consider i = 1, 2, ..., T. In the specific case of i = 8, panel 9 in Figure 3 shows that
σνlinear,8 = −.18 and σνnonlin,8 = −.42.
To these different baseline paths, we add the fiscal response in the first period the ZLB binds,
which happens in the same period as the adverse shock hits (t = 1). By letting{Slineart
(σlinearν,i , σG
)}Tt=1
and{Snonlint
(σnonlinν,i , σG
)}Tt=1
denote vectors with simulated variables in the linear and nonlinear
solutions when both the negative baseline shock σν and the positive government spending shock
σG hits the economy, we can compute the partial impact of the fiscal spending shock as
Ijt (ZLBDUR) = Sjt
(σjν,i, σG
)−Bjt
(σjν,i
)for j = {linear, nonlin} and where we write Ijt (ZLBDUR) to highlight its dependence on the
liquidity trap duration. Notice that the fiscal spending shock is the same for all i and is scaled so
15
-
that ZLBDUR remains the same as under the baseline shock only. By setting the fiscal impulse so
that the liquidity trap duration remains unaffected, we retrieve “marginal”spending multipliers in
the sense that they show the impact of a “tiny”change in the fiscal instrument.15
In Figure 4 we report the results of our exercise. The upper panels report results for the
benchmark calibration with the Kimball aggregator. The lower panels report results under the
Dixit-Stiglitz aggregator, in which case �p = 0. This parametrization implies a substantially higher
slope of the linearized Phillips curve (see eq. 21) and thus a much stronger sensitivity of expected
inflation to current and expected future marginal costs (and output gaps). We will first discuss the
results under the Kimball parameterization, and then turn to the Dixit-Stiglitz results.
The left panels report the output-spending multiplier on impact, i.e. simply
mi =∆Yt,i∆Gt,i
=∆Yt,i/Y
∆gy,t
where the ∆-operator represents the difference between the scenario with the spending change and
the baseline without the spending change and Y denotes the steady state level of output. We
compute mi for ZLBDUR = 1, ..., 12, but also include results for the case when the economy is at
the steady state, so that ZLBDUR = 0.
As the linear approximation is more accurate the closer the economy is to the steady state, it is
not surprising that the difference between the “linear”and “nonlinear”multiplier increases with the
duration of the liquidity trap. In a three-year liquidity trap, the multiplier in the nonlinear solution
is about twice as high (0.65) compared to normal times (when it is about 0.30), whereas it is about
7 times higher (2.1) in the linearized solution. So for a three year liquidity trap, the multiplier
in the linearized solution is over three times (2.1/0.65) as large as in the nonlinear solution. For
shorter-lived liquidity traps, the differences are notably more modest, and in the special case when
the economy is in the steady state (ZLBDUR = 0 in the figure) we note that the multipliers are
identical in both economies. The difference in government debt (as share of actual annualized
GDP) response after 1 year, shown in the upper right panel, largely follows the pattern for mi and
increases with ZLBDUR.16
The substantial differences in the output and debt responses between the linearized and nonlin-
ear solutions begs the question of which factors account for them. The middle upper panel, which
15 Had we considered a larger fiscal intervention that altered the duration of the liquidity trap, there would havebeen an important distinction between the average (i.e. the total response) and marginal (i.e. the impact of a smallchange in gt which leaves ZLBDUR unchanged) multiplier as discussed in further detail in Erceg and Linde (2014).16 For ease of interpretability, we have normalized the response of debt and inflation so that they correspond to a
initial change in government spending (as share of steady state output) by one percent.
16
-
shows the response of the one-period ahead expected annualized inflation rate (i.e., 4Etπt+1), sheds
some light on this. As can be seen from the panel, expected inflation responds much more in a
long-lived trap in the linearized model than in the nonlinear model. The sharp increase in ex-
pected inflation triggers a larger reduction in the actual real rate relative to the potential real
rate (not shown) in the linearized model, and thereby induces a more favorable response of private
consumption which helps to boost output relative to the nonlinear model.
Turning to the Dixit-Stiglitz case shown in the lower panels, we see that the differences between
the linear and nonlinear solutions are even more pronounced in this case, with the multiplier in
an 8-quarter trap being over 100 (27.5/0.25) times larger than in normal times in the linearized
solution, but only roughly 10 times higher in the nonlinear solution.17 The larger discrepancy
in the Dixit-Stiglitz case is to a large extent driven by the fact that we are in effect allowing a
substantially higher slope (i.e. κmc) of the New Keynesian Phillips curve in eq. (20). Taken
together, the results in Figure 4 suggest that the findings of the papers in the previous literature
which relied on linearized models were more distorted to the extent that they relied on a calibration
with a higher slope of the Phillips curve and thus a larger sensitivity of expected inflation.
The key question is then why expected inflation responds so much more in the linearized econ-
omy, and particularly so in the Dixit-Stiglitz case? To shed light on this, we simulate two additional
variants of the nonlinear model. In the first, we linearize the pricing equations of the model, e.g.
replace all pricing equations in the nonlinear model with the standard linearized Phillips curve.
In the second, we linearize all the pricing equations and remove the price distortion term from
the aggregate resource constraint (19). Following the approach with the linear and fully nonlinear
models, we construct baseline scenarios for the two additional variants of the model as described in
Section 3.1 for ZLBDUR = 1, ..., 12. The blue dash-dotted line in Figure 3 depicts the eight quarter
liquidity trap baseline in the variant with linearized pricing equations and resource constraint (sec-
ond additional variant described above). Clearly, the simulated paths of the variables in this variant
of the model are very similar to those in the fully linearized solution. Therefore, given that the
consumption demand νt generating the baseline and the added government spending shock both
work through the demand side of the economy, is it not surprising that the results in Figure 5 for
this model (blue dashed-dotted line, referred to as “Linearized Resource Constraint and NKPC”)
are very similar to those obtained with the linearized model, both under Kimball and Dixit-Stiglitz
aggregators. Hence, we can draw the conclusion that it is the linearization of the resource constraint17 We only show results up to 8 quarters with the Dixit-Stiglitz aggregator to be able to show the differences more
clearly in the graph.
17
-
and the Phillips curve (20), and not the aggregate demand part of the model, which account for
the bulk of the differences between in the linear and nonlinear models under Kimball in a long-lived
liquidity trap. In fact, as shown by the green dash-dotted line in the upper panels of Figure 5, it
is almost suffi cient to just linearize the NKPC to account for most of the discrepancy between the
linearized and nonlinear solution under the Kimball aggregator. Together, these findings show that
the lower inflation response in Figure 4 in the nonlinear solution relative to the linear solution is
driven by an effectively lower response to marginal costs when the price dispersion term is elevated.
The price dispersion term per se does not matter much, but the fact that the price dispersion is
elevated following an adverse shock implies that many firms percieve that their demand elasticity
is high and are therefore reluctant to change prices much at all in response to changes in marginal
costs (in terms of Figure 1, they are located in the upper left quadrant).
The lower panels in Figure 5, however, show that log-linearization of the New Keynesian Phillips
curve only is not suffi cient to explain the large discrepancies between the linear and nonlinear
solutions under the Dixit-Stiglitz aggregator. In this case, accounting for the price distortion in
the aggregate resource constraint is necessary, i.e. log-linearizing the resource constraint so that
movements in the price distortion term p∗t in eq. (18) becomes irrelevant for equilibrium dynamics.
The reason for this difference between the Kimball and the Dixit-Stiglitz aggregators is that the
price distortion variable moves much more for the latter specification as re-optimizing firms will
adjust their prices much more under Dixit-Stiglitz compared to Kimball for given ξp (recalling the
insights from Figure 1). So in a Dixit-Stiglitz world where firms adjust prices a lot when they
re-optimize, the bulk of the difference between the linearized and nonlinear solutions is driven by
movements in the price distortion, whereas in the Kimball world where firms adjust prices by less,
the bulk of the differences is driven by the pricing equations directly.
Boneva, Braun and Waki (2016) argue that the key is to account for the price distortion, and
our results suggest their claim is valid given that they are considering a model framework in line
with the Dixit-Stiglitz aggregator. In terms of multiplier sizes, it is important to note that we
report lower multipliers that Boneva, Braun and Waki in Figure 4 (for comparable degree of price
adjustment) because our spending process is assumed to be a fairly persistent AR(1) process. If
we assume that spending follows an uniform MA process and is only increased as long as policy
rates are constrained by the ZLB, we obtain a marginal multiplier of unity in both the linear and
nonlinear solutions already in a one-quarter liquidity trap (as we should, see Woodford, 2011, for
18
-
proof).18 Even so, there is an important difference between the linearized and nonlinear solution
for longer-lived liquidity traps, where multiplier rises to 5 in a three-year trap in the former case
but only to 1.03 in the nonlinear solution. This is in line with Boneva, Braun and Waki, who
reports a maximum multiplier of 1.05 in their model.
4. Robustness analysis
In this section, we examine the robustness of the results w.r.t. the aggregator specification (Kimball
vs. Dixit-Stiglitz) as well as to the indexation schedule for firms which are not able to re-optimize
their prices.
4.1. Dixit-Stiglitz vs. Kimball
To further tease out the difference between the Kimball vs. Dixit-Stiglitz aggregator, Panel A in
Figure 6 compares outcomes when the sticky price parameter ξp is adjusted in the Dixit-Stiglitz
version so that the slope of the linearized Phillips curve (20) is the same as in our benchmark
Kimball calibration. Both the Kimball and Dixit-Stiglitz versions hence now feature a linearized
Phillips curve with an identical slope coeffi cient (κmc = 0.012, see 21), but the Dixit-Stiglitz version
of the model achieves this with a substantially higher value of ξp (0.90). However, since only the
value of κmc matters in the linearized solution, the multiplier schedules are invariant w.r.t. the
mix of ξp and �p that achieves a given κmc in the linearized models. Consequently, the linearized
solution for the Dixit-Stiglitz aggregator is thus identical to the Kimball solution depicted by the
solid black line in the upper panel in Figure 4.
Even so, the nonlinear solutions shown in Panel A in Figure 6 differ. In particular, we see that
18 These results are discussed and presented in Appendix A, see Figure A.2.
19
-
the Dixit-Stiglitz aggregator implies that expected inflation and output multiplier responds more
when the duration of the liquidity trap increases. Thus, when the Kimball parameter �p is reduced,
the more will expected inflation and output multiplier respond when ZLBDUR increases; conversely,
increasing �p and lowering ξp flattens the output multiplier schedule even more. The explanation
behind this finding is that a higher value of �p induces the elasticity of demand to vary more with
the relative price differential among the intermediate good firms as shown in Figure 1, and this
price differential increases when the economy is far from the steady state. Thus, intermediate firms
which only infrequently are able to re-optimize their price will optimally choose to respond less to
a given fiscal impetus far from the steady state when price differentials are larger as they perceive
that they may have a much larger impact on their demand for a given change in their relative price.
As a result, aggregate current and expected inflation are less affected far from the steady state in the
Kimball case relative to the Dixit-Stiglitz case for which the elasticity of demand is independent of
relative price differentials. This demonstrates that the modeling of price frictions matters
importantly within a nonlinear framework, especially so when nominal wages are flexible.
4.2. Indexation of non-optimizing firms
So far, we have followed the convention in the literature and assumed that non-optimizing firms
index their prices w.r.t. the steady state inflation rate, see eq. (12). This is a convenient benchmark
modelling assumption as it simplifies the analysis by removing steady state price distortions. How-
ever, this assumption have been criticized for being inconsistent with the microevidence on price
setting. According to micro evidence on price setting, many firms’prices remain unchanged for
several subsequent quarters, whereas whey always change under our benchmark indexation scheme.
Thus —there is an important issue to what extent this matters for aggregate dynamics, especially
in the nonlinear solution. To examine this, we re-specify the model assuming no indexation among
the non-optimizing firms following e.g. Ascari and Ropele (2007), i.e.
P̃t = Pt−1. (23)
In Panel B in Figure 6, we report the results when comparing the nonlinear baseline model
(black solid line, which features indexation) with the nonlinear variant without indexation for the
non-optimizing firms (red dotted line). From the panels, we see that abandoning the conventional
assumption of full indexation results in a somewhat steeper multiplier schedule, mostly explained
by the somewhat higher sensitivity of expected inflation in the “no-indexation” model. As the
20
-
marginal multipliers in the linearized solution is roughly unchanged (verified, but not shown in the
figure), the message in Panel B is that the results by and large hold up well for this perturbation
of the model as well.
5. Conclusions
We have calculated the magnitude of the fiscal spending multiplier in linearized and nonlinear
solutions of a New Keynesian model at the zero lower bound. Importantly, we use a model that is
amended with real rigidities to simultaneously account for the macroeconomic evidence of a low
Phillips curve slope and the microeconomic evidence of frequent price re-optimization. We have
shown that the nonlinear solution is associated with a much smaller multiplier than the linearized
solution in long-lived liquidity traps, and pinned down the key features in the model which account
for the difference. Our results caution against the common practice of using linearized models to
calculate fiscal multipliers in long-lived liquidity traps.
21
-
References
Adam, Klaus, and Roberto M. Billi (2006), “Optimal Monetary Policy under Commitment witha Zero Bound on Nominal Interest Rates”, Journal of Money, Credit, and Banking 38(7),1877-1905.
Adam, Klaus, and Roberto M. Billi (2007), “Discretionary Monetary Policy and the Zero LowerBound on Nominal Interest Rates”, Journal of Monetary Economics 54(3), 728-752.
Adolfson, Malin, Stefan Laséen, Jesper Lindé and Mattias Villani (2005), “The Role of StickyPrices in an Open Economy DSGEModel: A Bayesian Investigation”, Journal of the EuropeanEconomic Association Papers and Proceedings 3(2-3), 444-457.
Altig, David, Christiano, Lawrence J., Eichenbaum, Martin and Jesper Lindé (2011), “Firm-SpecificCapital, Nominal Rigidities and the Business Cycle”, Review of Economic Dynamics 14(2),225-247.
Ascari, Guido and Tiziano Ropele (2007) “Optimal Monetary Policy Under Low Trend Inflation,”Journal of Monetary Economics 54, 2568-2583
Bernanke, Ben, Gertler, Mark and Simon Gilchrist (1999), “The Financial Accelerator in a Quan-titative Business Cycle Framework”, in John B. Taylor and Michael Woodford (Eds.), Hand-book of Macroeconomics, North-Holland Elsevier Science, New York.
Bi, Huixin, Eric Leeper, and Campbell Leith (2011), "Uncertain Fiscal Consolidations." Manu-script.
Blanchard, Olivier and Roberto Perotti (2002), “An Empirical Characterization of The DynamicEffects of Changes in Government Spending and Taxes on Output”, The Quarterly Journalof Economics 117(4), 1329-1368.
Blanchard, Olivier, Christopher Erceg and Jesper Lindé (2016), “Jump-Starting the Euro Area Re-covery: Would a Rise in Core Fiscal Spending Help the Periphery?”, NBER MacroeconomicsAnnual 31, forthcoming.
Boneva, Lena Mareen, Braun, R. Anton and Yuichiro Waki (2016), “Some unpleasant propertiesof loglinearized solutions when the nominal rate is zero,”Journal of Monetary Economics 84,216-232.
Braun, R. Anton, Lena Mareen Kvrber and Yuichiro Waki (2013), “Small and orthodox fiscal mul-tipliers at the zero lower bound,”Working Paper 2013-13, Federal Reserve Bank of Atlanta.
Christiano, Lawrence J., Martin Eichenbaum and Charles Evans (2005), “Nominal Rigidities andthe Dynamic Effects of a Shock to Monetary Policy”, Journal of Political Economy 113(1),1-45.
Christiano, Lawrence, Martin Eichenbaum and Benjamin K. Johannsen (2016), “Does the NewKeynesian Model Have a Uniqueness Problem?”, manuscript, Northwestern University.
Christiano, Lawrence, Martin Eichenbaum and Sergio Rebelo (2011), “When is the GovernmentSpending Multiplier Large?”Journal of Political Economy 119(1), 78-121.
Christiano, Lawrence, Motto, Roberto and Massimo Rostagno (2008), “Shocks, Structures or Mon-etary Policies? The Euro Area and the US After 2001”, Journal of Economic Dynamics andControl 32(8), 2476-2506.
Christiano, Lawrence J., Mathias Trabandt and Karl Walentin (2007), “Introducing financial fric-tions and unemployment into a small open economy model,” Sveriges Riksbank WorkingPaper Series No. 214.
22
-
Christiano, Lawrence J., Mathias Trabandt and Karl Walentin (2011), “DSGEModels for MonetaryPolicy Analysis,”chapter 7 in: Benjamin M. Friedman & Michael Woodford (eds.), Handbookof Monetary Economics, pp. 285-367 Elsevier.
Coenen, Günter, Christopher Erceg, Charles Freedman, Davide Furceri, Michael Kumhof, RenéLalonde, Douglas Laxton, Jesper Lindé, Annabelle Mourougane, Dirk Muir, Susanna Mursula,Carlos de Resende, John Roberts, Werner Roeger, Stephen Snudden, Mathias Trabandt andJan in ’t Veld (2012), “Effects of Fiscal Stimulus in Structural Models”, American EconomicJournal: Macroeconomics 4(1), 22-68.
Cogan, John F., Cwik, Tobias, Taylor, John B. and Volker Wieland (2010), “New Keynesianversus Old Keynesian Government Spending Multipliers”, Journal of Economic Dynamicsand Control 34, 281-95.
Davig, Troy and Eric M. Leeper (2009), “Monetary-Fiscal Policy Interactions and Fiscal Stimulus,”European Economic Review 55(2), 211-227.
Dixit, Avinash K, and Joseph E. Stiglitz (1977), “Monopolistic Competition and Optimum ProductDiversity”, American Economic Review 67(3), 297-308.
Drautzburg, Thorsten and Harald Uhlig (2011), “Fiscal Stimulus and Distortionary Taxation”,NBER Working Papers No. 17111, National Bureau of Economic Research.
Dupor, Bill and Rong Li (2015), “The Expected Inflation Channel of Government Spending in thePostwar U.S.,”European Economic Review 74(1), 36-56.
Eggertsson, Gauti and Michael Woodford (2003), “The Zero Interest-Rate Bound and OptimalMonetary Policy”, Brookings Papers on Economic Activity 1, 139-211.
Eggertsson, Gauti (2008), “Great Expectations and the End of the Depression”, American Eco-nomic Review 98(4), 1476-1516.
Eggertsson, Gauti (2010), “What Fiscal Policy Is Effective at Zero Interest Rates?”, NBER Macro-economics Annual 25, 59-112.
Eggertsson, Gauti and Sanjay R. Singh (2016), “Log-linear approximation versus an exact solutionat the ZLB in the New Keynesian model”, manuscript, Brown University.
Erceg, Christopher, Guerrieri, Luca and Christopher Gust (2006), “SIGMA: A New Open EconomyModel for Policy Analysis”, Journal of International Central Banking 2(1), 1-50.
Erceg, Christopher and Jesper Lindé (2014), “Is There A Fiscal Free Lunch In a Liquidity Trap?”,Journal of the European Economic Association 12(1), 73-107.
Fair, Ray C. and John B. Taylor (1983), “Solution and Maximum Likelihood Estimation of DynamicRational Expectations Models”, Econometrica 51(4), 1169-1185.
Fernández-Villaverde, Jesús, Gordon, Grey, Guerrón-Quintana, Pablo and Juan F. Rubio-Ramírez(2015), “Nonlinear Adventures at the Zero Lower Bound”, Journal of Economic Dynamics &Control 57, 182—204.
Galí, Jordi and Mark (1999), “Inflation Dynamics: A Structural Econometric Analysis”, Journalof Monetary Economics, 44, 195-220.
Galí, Jordi, Gertler, Mark and David López-Salido (2001), “European Inflation Dynamics”, Euro-pean Economic Review, 45, 1237-70.
Galí, Jordi, López-Salido, David and Javier Vallés (2007), “Understanding the Effects of Gov-ernment Spending on Consumption”, Journal of the European Economic Association, 5(1),227-270.
23
-
Gust, Christopher J., Edward P. Herbst, J. David Lopez-Salido, and Matthew E. Smith (2016).“The Empirical Implications of the Interest-Rate Lower Bound,”Board of Governors of theFederal Reserve System Finance and Economics Discussion Series 2012-83r.
Hall, Robert E. (2009), “By How Much Does GDP Rise if the Government Buys More Output?”,Brookings Papers on Economic Activity 2, 183-231.
Hebden, J.S., Lindé, J., Svensson, L.E.O., 2009. Optimal monetary policy in the hybrid newKeynesian model under the zero lower bound constraint. Mimeo, Federal Reserve Board.
Judd, Kenneth L., Maliar, Lilia and Serguei Maliar (2011), “Numerically stable and accurate sto-chastic simulation approaches for solving dynamic economic models,”Quantitative Economics2(2), 173-210.
Juillard, Michel (1996), “Dynare : A Program for the Resolution and Simulation of DynamicModels with Forward Variables Through the Use of a Relaxation Algorithm,”CEPREMAPWorking Paper 9602.
Jung, Taehun, Yuki Teranishi, and Tsotumu Watanabe (2005), “Optimal Monetary Policy at theZero-Interest-Rate Bound," Journal of Money, Credit, and Banking, 37 (5), 813-35.
Kimball, Miles S. (1995), “The Quantitative Analytics of the Basic Neomonetarist Model,”Journalof Money, Credit, and Banking 27(4), 1241—1277.
Klenow, Peter J. and Benjamin A. Malin (2010), “Microeconomic Evidence on Price-Setting”,Chapter 6 in Benjamin M. Friedman and Michael Woodford (Eds.), Handbook of MonetaryEconomics, Elsevier, New York.
Keynes, John Maynard (1936), The General Theory of Employment, Interest and Money, London:Macmillan Press.
Leeper, Eric M., Nora Traum, Todd B. Walker (2015), “Cleaning Up the Fiscal Multiplier Morass”,manuscript, Indiana and North Carolina State Universities.
Lindé, Jesper (2005), “Estimating New Keynesian Phillips Curves: A Full Information MaximumLikelihood Approach”, Journal of Monetary Economics, 52(6), 1135-49.
Mertens, Karel and Morten Ravn (2010), “Fiscal Policy in an Expectations Driven Liquidity Trap”,CEPR Discussion Paper No. 7931.
Michaillat, Pascal (2014), “A Theory of Countercyclical Government Multiplier”, American Eco-nomic Journal: Macroeconomics 6(1), 190-217.
Nakata, Taisuke (2015), “Uncertainty at the Zero Lower Bound”, American Economic Journal:Macroeconomics forthcoming.
Perotti, Roberto (2007), “In Search of the Transmission Mechanism of Fiscal Policy”, NBER Macro-economics Annual 22, 169-226.
Ramey, Valerie A. (2011), “Identifying Government Spending Shocks: It’s All in the Timing”,Quarterly Journal Of Economics 126(1), 1-50.
Rendahl, Pontus (2016), “Fiscal Policy in an Unemployment Crisis”, Review of Economic Studies83(3), 1189-1224.
Roulleau-Pasdeloup, Jordan (2016), “The Government Spending Multiplier in a Deep Recession”,manuscript, University of Lausanne.
Smets, Frank and Raf Wouters (2003), “An Estimated Stochastic Dynamic General EquilibriumModel of the Euro Area”, Journal of the European Economic Association 1(5), 1123-1175.
24
-
Smets, Frank and Raf Wouters (2007), “Shocks and Frictions in US Business Cycles: A BayesianDSGE Approach”, American Economic Review 97(3), 586-606.
Traum, Nora and Shu-Chun S. Yang (2011), “Monetary and Fiscal Policy Interactions in the Post-War U.S.”, European Economic Review 55(1), 140-164.
Uhlig, Harald (2010), “Some Fiscal Calculus”, American Economic Review Papers and Proceedings100(2), 30-34.
Woodford, Michael, 2011, “Simple Analytics of the Government Spending Multiplier", AmericanEconomic Journal: Macroeconomics 3(1), 1-35.
25
-
Notes: This figure is taken from Levin, Lopez-Salido and Yun, 2007, "Strategic Complementarities and Optimal Monetary Policy", CEPR Discussion Paper No. 6423
-
0 5 10 15 20
−2.5
−2
−1.5
−1
−0.5
0
Per
cent
1. Output Gap
0 5 10 15 20
−2
−1
0
1
Per
cent
2. Yearly Inflation (ln(Pt/P
t−4))
0 5 10 15 200
1
2
3
4
Per
cent
3. Nominal Interest Rate (APR)
0 5 10 15 20
0
0.5
1
1.5
2
2.5
Per
cent
4. Real Interest Rate (APR)
0 5 10 15 20
−0.5
0
0.5
1
1.5
2
Per
cent
5. Potential Real Interest Rate (APR)
0 5 10 15 200
0.05
0.1
Per
cent
6. Price Dispersion
0 5 10 15 20
−6
−4
−2
0
Per
cent
Quarters
7. Real GDP
0 5 10 15 2060
62
64
66
Per
cent
Quarters
8. Government Debt to GDP
Figure 2: Baselines in Linear and Nonlinear Models for an Equally−Sized Consumption Demand Shock
0 5 10 15 20
−0.15
−0.1
−0.05
0
Per
cent
Quarters
9. Consumption Demand Shock
Linear Model Nonlinear Model
-
0 5 10 15 20−10
−8
−6
−4
−2
0
Per
cent
1. Output Gap
0 5 10 15 20
−2
−1
0
1
Per
cent
2. Yearly Inflation (ln(Pt/P
t−4))
0 5 10 15 200
1
2
3
4
Per
cent
3. Nominal Interest Rate (APR)
0 5 10 15 20
0
1
2
Per
cent
4. Real Interest Rate (APR)
0 5 10 15 20−3
−2
−1
0
1
2
Per
cent
5. Potential Real Interest Rate (APR)
0 5 10 15 200
0.05
0.1
0.15
0.2
0.25
Per
cent
6. Price Dispersion
0 5 10 15 20
−15
−10
−5
0
Per
cent
Quarters
7. Real GDP
0 5 10 15 2060
65
70
75
80
Per
cent
Quarters
8. Government Debt to GDP
Figure 3: Baselines for 8−Quarter Liquidity Trap
0 5 10 15 20−0.4
−0.3
−0.2
−0.1
0
Quarters
Per
cent
9. Consumption Demand Shock
Linear Model Nonlinear Model Nonlinear Model with Linear NKPC and Res . Con.
-
0 2 4 6 8 10 12
0.5
1
1.5
2
Impact Spending Multiplier
ZLB duration, quarters
Mul
tiplie
r
0 2 4 6 8 10 120
0.2
0.4
0.6
0.8
1
1.2
1.4
1.6
Expected inflation (4⋅Etπt+1)
ZLB duration, quartersP
erce
ntag
e P
oint
s0 2 4 6 8 10 12
-6
-4
-2
0
2
Govt Debt to GDP (After 1 Year)
ZLB duration, quarters
Per
cent
age
Poi
nts
0 2 4 6 80
5
10
15
20
25
30Impact Spending Multiplier
ZLB duration, quarters
Mul
tiplie
r
0 2 4 6 80
10
20
30
40
50
60
70
80Expected Inflation (4⋅Etπt+1)
ZLB duration, quarters
Per
cent
age
Poi
nts
Figure 4: Marginal Multipliers
0 2 4 6 8
-200
-150
-100
-50
0
Govt Debt to GDP (After 1 Year)
ZLB duration, quarters
Per
cent
age
Poi
nts
Linear Model Nonlinear Model
Benchmark Calibration
Alternative Calibration: Dixit Stiglitz
-
0 2 4 6 8 10 12
0.5
1
1.5
2
Impact Spending Multiplier
ZLB duration, quarters
Mul
tiplie
r
0 2 4 6 8 10 120
0.2
0.4
0.6
0.8
1
1.2
1.4
1.6
Expected inflation (4⋅Etπt+1)
ZLB duration, quartersP
erce
ntag
e P
oint
s0 2 4 6 8 10 12
-6
-4
-2
0
2
Govt Debt to GDP (After 1 Year)
ZLB duration, quarters
Per
cent
age
Poi
nts
0 2 4 6 80
5
10
15
20
25
30Impact Spending Multiplier
ZLB duration, quarters
Mul
tiplie
r
0 2 4 6 80
10
20
30
40
50
60
70
80Expected Inflation (4⋅Etπt+1)
ZLB duration, quarters
Per
cent
age
Poi
nts
Figure 5: Marginal Multipliers
0 2 4 6 8
-200
-150
-100
-50
0
Govt Debt to GDP (After 1 Year)
ZLB duration, quarters
Per
cent
age
Poi
nts
Linear Model Nonlinear Model Linearized Resource Constraint and NKPC Linearized NKPC only
Benchmark Calibration
Alternative Calibration: Dixit Stiglitz
-
0 2 4 6 8 10 12
0.5
1
1.5
2
Impact Spending Multiplier
ZLB duration, quarters
Mul
tiplie
r
0 2 4 6 8 10 120
0.5
1
1.5
Expected Inflation (4Ett+1)
Per
cent
age
Poi
nts
ZLB duration, quarters0 2 4 6 8 10 12
-6
-4
-2
0
2
Govt Debt to GDP (After 1 Year)
ZLB duration, quarters
Per
cent
age
Poi
nts
Kimball (Benchmark) Dixit Stiglitz
0 2 4 6 8 10 12
0.5
1
1.5
2
Impact Spending Multiplier
ZLB duration, quarters
Mul
tiplie
r
0 2 4 6 8 10 120
0.5
1
1.5
Expected inflation (4Ett+1)
ZLB duration, quarters
Per
cent
age
Poi
nts
Figure 6: Marginal Impact of Changes in Spending: Sensitivity Analysis in Nonlinear Model
0 2 4 6 8 10 12
-6
-4
-2
0
2
Govt Debt to GDP (After 1 Year)
ZLB duration, quarters
Per
cent
age
Poi
nts
With Indexation (Benchmark) No Indexation
Panel A: Kimball (p=0.667; p>0) vs. Dixit-Stiglitz (p=0.9; p=0)
Panel B: Impact of Indexation Assumption for Non-Optimizing Firms
-
Appendix A. Additional Results
In this appendix, we state the log-linearized equations of the model and present some additional
results.
A.1. The Log-linearized Stylized Model
As shown in the technical appendix (available upon request from the authors), the equations of the
log-linearized model can be written as follows:
xt = xt+1|t − σ̂(it − πt+1|t − rpott ), (A.1)
πt = βπt+1|t + κpxt, (A.2)
ypott =1
φmcσ̂[gy,t + (1− gy)ννt], (A.3)
rpott = −δt +1
σ̂
(1− 1
φmcσ̂
)[(gy,t − gy,t+1|t) + (1− gy)ν(νt − νt+1|t)
], (A.4)
bG,t = (1 + r)bG,t−1 + bG(it−1 − πt) + gy,t − τNsN (yt + φmcxt)− τ t, (A.5)
yt = xt + ypott (A.6)
where σ̂, κp, φmc and sN are composite parameters defined as:
σ̂ = σ(1− gy)(1− νc), (A.7)
κp =(1− ξp)(1− βξp)
ξp
1
1 +(φp − 1
)�pφmc, (A.8)
φmc =χ
1− α +1
σ̂+
α
1− α, (A.9)
sN =1− αφp
. (A.10)
In slight abuse of previous notation, all variables above are measured as percent or percentage
point deviations from their steady state level.A.1
A.1 We use the notation yt+j|t to denote the conditional expectation of a variable y at period t + j based oninformation available at t, i.e., yt+j|t = Etyt+j . The superscript ‘pot’ denotes the level of a variable that wouldprevail under completely flexible prices, e.g., ypott is potential output.
32
-
Equation (A.1) expresses the “New Keynesian”IS curve in terms of the output and real interest
rate gaps. Thus, the output gap xt depends inversely on the deviation of the real interest rate
(it − πt+1|t) from its potential rate rpott , as well as on the expected output gap in the following
period. The parameter σ̂ determines the sensitivity of the output gap to the real interest rate;
as indicated by (A.7), it depends on the household’s intertemporal elasticity of substitution in
consumption σ, the steady state government spending share of output gy, and a (small) adjustment
factor νc which scales the consumption taste shock νt. The price-setting equation (A.2) specifies
current inflation πt to depend on expected inflation and the output gap, where the sensitivity to
the latter is determined by the composite parameter κp. Given the Calvo-Yun contract structure,
equation (A.8) implies that κp varies directly with the sensitivity of marginal cost to the output
gap φmc, and inversely with the mean contract duration (1
1−ξp). The marginal cost sensitivity
equals the sum of the absolute value of the slopes of the labor supply and labor demand schedules
that would prevail under flexible prices: accordingly, as seen in (A.9), φmc varies inversely with the
Frisch elasticity of labor supply 1χ , the interest-sensitivity of aggregate demand σ̂, and the labor
share in production (1− α). The equations (A.3) and (A.4) determinate potential output and the
potential (or natural) real rate. The evolution of government debt is determined by equation (A.5) ,
and depends on variations in the service cost of debt, government spending as well as labor income
and lump-sum tax revenues. Equation (A.6) is a simple definitional equation for actual output yt
(in logs). Finally, the policy rate it follows a Taylor rule subject to the zero lower bound (equation
16 in the main text) and the exogenous shocks follows the processes in eqs. (22).
A.2. Sensitivity Analysis
Figure A.1 report results when the discount factor shock δt (defined as the expected change in ςt
shock in eq. 1, see eq. 4) is driving the baseline in Figure 3. For ease of comparison, the benchmark
results with the consumption taste shock νt driving the baseline scenarios are also included. The
upper panels in the figure confirm the results in by Erceg and Lindé (2014) by showing that the fiscal
spending multiplier is independent of the shock driving the baseline when the model is linearized
as long as the different baseline shocks generate an equally long-lived ZLB episode. So our choice
to work with the consumption demand shock in νt+j instead of the conventional discount factor
shock in ςt in (1) to generate the baseline path underlying Figures 4 to 6 has no consequence for
our results with the linearized model. However, the results for the non-linear variant may differ.
However, the lower panels in Figure A.1 show that the results are very similar even in the nonlinear
33
-
solution, so our choice of baseline appears immaterial.
Another aspect we want to understand is how our results differ from Boneva, Braun, and Waki
(2016) due to our AR(1) assumption for government spending instead of the MA-process they work
with. Figure A.2 assess this issue by comparing the results under our benchmark AR(1) process
with persistence .95 for government spending against the MA process for which Gt is increased
in an uniform fashion as long at the policy rate is bounded at zero for ZLBDUR = 1, 2, 3, ..., T
and set at its steady state value otherwise. Apart from the fact that our solution procedure does
not account for future shock uncertainty, this way of modeling government spending is identical to
Boneva et al. who in turns follow Eggertsson (2010).
As can be seen from the upper panels of Figure A.2, the MA-process increases the marginal
spending multiplier substantially relative to the AR(1) process for the ZLB durations we consider.
This happens as increases in government spending has very benign effects on the potential real
interest rate when the duration of the spending hike equals the expected duration of the liquidity
trap (see e.g. Erceg and Lindé, 2014). For a one quarter liquidity trap the multiplier equals unity,
as shown analytically by Woodford (2011). Our fairly persistent AR(1) process tends to dampen
the multiplier schedule as a relatively large fraction of the spending comes on line when the ZLB
is no longer binding. This feature explains why the AR(1) multiplier is substantially lower in a
short lived liquidity trap. However, the AR(1) process is also associated with a substantially lower
multiplier even in a fairly long-lived trap compared to the MA process because its has less benign
effects on the potential real rate.
All this is well-known from the literature on linearized models. However, the results for the non-
linear model, shown in the lower panels, are less explored. We have already discussed the AR(1)
case at length in the text. What we see is that the results for the MA process are quite different
for longer ZLB durations, because the MA schedule for the nonlinear model stays essentially flat
at unity, in line with the findings of Braun et al. (2013); for a 12-quarter trap the multiplier only
increases to 1.03 from a multiplier of unity in a one-period liquidity trap. This is sharp contrast to
the multiplier schedule for the linearized model where the multiplier is as high as 5 in a liquidity
trap lasting 3 years. All told, the results in this appendix shows that our benchmark results holds
up well for an MA-process for government spending. If anything, an MA process magnifies the
difference between the linearized and nonlinear solutions. Moreover, the results for the linear and
nonlinear models in Figure A.2 are in line with the results in the existing literature.
34
-
0 2 4 6 8 10 12
0.5
1
1.5
2
Impact Spending Multiplier
ZLB duration, quarters
Mul
tiplie
r
0 2 4 6 8 10 120
0.2
0.4
0.6
0.8
1